Contribution
It names the assumption the comment rests on and gives data against it. A synchronous-minus-asynchronous difference controls for everything except timing only if participants expect the same from both conditions, and in the target sample they did not. The reply adds that the expectancy difference did not vary with hypnotisability, which turns Ehrsson et al.’s central null from evidence against the demand account into something the account predicts. It also separates the ownership item from the two referred-touch items and shows ownership is the weakest and the most dependent on hypnotisability.
Key insight
A control condition protects an inference only against what it holds constant. The asynchronous condition holds the stimuli, the room and the questions constant, but not what participants think should happen. If everyone expects more from synchrony, by the same amount, then everyone has the same reason to report more under synchrony, and a trait that governs how they meet the demand, by experience or by bias, need not change how much the two conditions differ. On this reading the synchrony effect is real as a difference in report and silent about its cause.
Assumptions
- Expectancy measures capture demand. Participants’ predicted experience for each condition, collected in the target study, is taken as the demand the conditions convey (Orne’s “totality of cues which convey an experimental hypothesis to the subject”, p. 1).
- Two routes to one report. “Participants can respond to the differing demand characteristics by either generating the corresponding experiences (if they have high trait capacity for phenomenological control, i.e. hypnotisability) or by response bias (if they have low capacity for phenomenological control)” (p. 1). The second route is proposed, not measured.
- Raw slopes over standardised correlations. The authors interpret linear models in scale units, and dispute that R² = 0.09 is weak.
- Instructions that did not move expectancy leave the question open. The target study’s suggestion manipulation is treated as a failed manipulation, so the comment’s null across instruction groups is not evidence about suggestion.
Key results
- The frame (p. 1). “All but one [reanalysis] confirm the findings we presented in the target paper, and the sole new analysis is insensitive and therefore uninformative. The disagreement is therefore not about data or analyses, but interpretation.” The insensitive one is the whole-sample drift analysis. “The asynchronous condition offers no protection against demand characteristic effects (including faking, imagination and phenomenological control).”
- Expectancies (p. 1). “Participant expectancies are greater for the synchronous than asynchronous condition”, as reported in the target article and elsewhere. In the target data (n = 353), hypnotisability does not predict the expectancy difference: b = −0.16 Likert units per SWASH unit, SE = 0.09, t = 1.78, P = 0.072, BH(0,0.25) = 0.07, rs = −0.08, 95% CI [−0.18, 0.03]. A B of 0.07 is evidence for no relation. “Participant expectancies arising from demand characteristics readily account for our reported null result, since these expectancies do not vary with the level of hypnotisability.” The size of the expectancy difference itself is not given in the reply.
- Implicit measures (p. 1). “This applies equally to implicit measures of the RHI (e.g., skin conductance response and proprioceptive drift) … people expect the patterns of results that are typically obtained in RHI experiments”, citing Lush, Seth & Dienes 2021, not read here.
- Effect size (pp. 1–2). Against “weak”: Cohen calls R² = 0.09 a medium correlation, and Funder & Ozer “a large effect that is potentially powerful in both the short and the long run”. The target article’s linear model shows “a 0.6 increase in mean score for the three illusion statements (7-point scale) for each 1 point increase in SWASH (5-point scale)”. This is for the synchronous condition.
- Ownership against referred touch (p. 2, Fig. 1). Mean agreement (maximum 3): S1 1.9, SE 0.1; S2 1.2, SE 0.1; S3, ownership, 0.7, with SE printed as 1.0 in both the PDF and the JATS text, probably a typo for 0.1 (my reading). Regressing ownership agreement on SWASH: b = 0.76, SE = 0.13, t = 5.79, P < 0.001, 95% CI [0.50, 1.01], BH(0,1.4) = 4.2 × 10⁶, rs = 0.27. “Minimum ‘ownership’ agreement (a score of 1) is predicted only for SWASH scores greater than two (the top 31% of hypnotisability scores)”. The caption does not say which condition’s ratings are regressed; the text’s framing implies the synchronous one.
- Order effects (p. 2). In a reanalysis reported in a preprint (Lush 2021, not read), “mean agreement with the ownership statement is seen only in the group who underwent the asynchronous ‘control’ task first”.
- The quartiles (p. 2). The residual illusion at low hypnotisability is already visible in the target article’s linear model; it “may reflect other effects arising from demand characteristics (e.g., bias or compliance effects)”. “For the crucial ownership statement, there may not be a residual effect (Fig. 1).”
- The literature review (p. 2). The commonest method in the 20 most influential RHI papers “is to test the difference between synchronous and asynchronous conditions, and then interpret only the synchronous condition”. Even if less cited papers used difference measures more, “difference measures are confounded by expectancies”.
- The remaining points (p. 2). 2AFC methods are no more resistant to demand “than the more common methods”. Similar correlations for synchronous and asynchronous conditions are explained if “both reflect suggestion effects”; a lower score for the visual-hallucination item may reflect lower expectancy or greater difficulty. The comment’s analyses are “also exploratory and post-hoc”.
- Conclusion (p. 2). “Future experiments, with adequate control conditions, are needed to establish whether mechanisms beyond demand characteristic effects (e.g., multisensory integration) play a significant role in RHI measures.”
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | The comment’s reanalyses agree with the target article’s results; the dispute is over interpretation | strong | the comment’s numbers are not contested; the drift reanalysis is inconclusive by its own Bayes factor |
| C2 | The asynchronous condition does not control for demand characteristics, because expectancies are higher for the synchronous condition | moderate | expectancy data in the target study and cited work, not given numerically here |
| C3 | The expectancy difference does not vary with hypnotisability | moderate | n = 353, b = −0.16, P = 0.072, B = 0.07; the slope’s sign and the t value as printed do not match |
| C4 | The null relation between hypnotisability and the synchrony effect is therefore predicted by the demand account | moderate as an inference | follows from C2 and C3 together with the two-routes assumption |
| C5 | High-hypnotisable participants meet demand by generating experience, low-hypnotisable ones by response bias | weak | proposed; neither route is measured apart from report |
| C6 | Agreement with the ownership item rises with hypnotisability and is, on average, minimal outside the top third | moderate | b = 0.76, rs = 0.27, P < 0.001; condition not stated in the caption; post hoc |
| C7 | Implicit and forced-choice RHI measures are open to the same confound | weak here | cited for drift and skin conductance, not read; asserted for 2AFC and cross-modal congruency, “not been tested” |
| C8 | Whether multisensory integration plays a significant role in RHI measures is not yet established | moderate as a statement of what the controls show | follows from C2 if the confound is granted; it is a claim about evidence, not that integration plays no role |
Concepts
- Demand characteristics — Orne’s sense: the cues in an experiment that convey its hypothesis to the participant.
- Phenomenological control — trait capacity to generate experience that meets expectation, measured as hypnotisability (Dienes et al.’s “cold control”, cited).
- Expectancies — participants’ predicted ratings for each condition, the reply’s operational measure of demand.
- Referred touch against ownership — S1 and S2 against S3, which the reply treats separately, following Reader, Trifonova & Ehrsson 2021, an Ehrsson paper.
Connections
The reply’s evidence lives mostly in works filed or cited elsewhere: the target article, Lush et al. 2020; Lush’s 2020 Collabra paper on demand characteristics; Roseboom & Lush on interpreting rubber-hand experiments (cited as a 2020 preprint; Roseboom & Lush 2022 is being filed by a sibling); and Lush, Seth & Dienes 2021 on expectancies for implicit measures, not held. It answers Ehrsson et al. 2022 (LIT-1264, NOTE-997), point by point except one: the graded temporal window of asynchrony, the comment’s strongest argument, is not taken up by name (the reply says remaining points are answered “briefly here, or in the original paper, or in our earlier preprint”).
Who is on which side. Anil Seth, second author here, wrote Seth 2013 (LIT-1263, NOTE-1000), which reported the cardio-visual rubber hand of his own group as evidence that interoceptive signals enter body ownership. Manos Tsakiris, of Apps & Tsakiris 2014 (LIT-1262, NOTE-991), signs the comment. The record’s two predictive accounts of ownership, THEORY-769 and THEORY-767, thus have source authors on opposite sides of whether the rubber-hand data show multisensory inference at all.
Bearing on the record
Read against the comment, NOTE-997. Where the two NOTEs touch the same THEORY, the verdict is this one’s: the reply is the half that changes it.
- THEORY-765 (Proposed), the synchrony effect. The exchange together partly meets its promote_when and shows a flaw in it. The comment meets the disjunct “trait suggestibility is measured and the synchrony effect survives it” on the questionnaire, at n = 353. This reply shows that the other disjunct, expectations “measured or matched”, was measured in the same sample and failed: expectations were not matched, and the mismatch is uniform across suggestibility, so survival of the effect does not discriminate the multisensory reading from the demand reading. The THEORY’s own “What this does not say” already lists “It does not separate synchrony from expectation”; after this exchange that is still true, and now shown on a large sample. Neither refutation condition is met: no design here matches expectations, and drift under asynchrony is not reported. The synchrony effect as a difference in report is, if anything, better supported than before: both sides accept it in 353 participants at every level of suggestibility. Proposed change (proposal only): drop “or in which trait suggestibility is measured and the synchrony effect survives it” as a sufficient condition, and require that expectations be matched across conditions, or measured and the synchrony effect shown to exceed what they predict; and add that a first-hand reading of Lush et al. 2020 itself is needed for the replication’s sample sizes and effect sizes, which neither of these papers gives for the synchrony contrast. The status should stay Proposed.
- THEORY-769 (Proposed), graded common-cause ownership. Bears against its readiness, not its content. Its promote_when asks for a parametric causal-inference test; the reply’s claim that 2AFC and other measures are open to the same demand would apply to such a test unless expectations are controlled, and the comment’s temporal-window studies, the closest thing to it, are cited by one side and passed over by the other. The ownership result is relevant: the item the THEORY is about is the weakest of the three on average and rises steeply with hypnotisability in the reply’s Figure 1, while its synchrony effect is unrelated to hypnotisability (comment, Suppl. Fig. 1C). Graded, yes, but graded across people by a trait the THEORY does not name. Proposed change: add to promote_when that the model-based test hold expectations constant across the conditions it varies, or measure them as a covariate.
- THEORY-767 (Proposed), interoception. Bears against the evidence it rests on, by the reply’s logic and not by its words. The cardio-visual hand of Suzuki et al. (Seth’s group) is a synchronous against asynchronous comparison on questionnaire and drift, which is the design this reply says “offers no protection against demand characteristic effects”. The reply does not mention that study; the application is mine. The sensitivity–susceptibility leg (Tsakiris et al. 2011) is also exposed: if susceptibility partly measures phenomenological control, a correlation with heartbeat detection could run through it. Proposed change: add to promote_when that expectancies be matched or measured across the cardio-visual conditions, and that the sensitivity–susceptibility relation survive controlling for hypnotisability. The non-bodily periodic control the THEORY already asks for does not answer this, since participants could expect less from it too.
- THEORY-768 (Proposed), minimal selfhood. No direct bearing: neither paper measures self-location or whole-body identification. The full-body illusion it leans on (Lenggenhager et al. 2007, LIT-1279) uses the same synchronous-against-asynchronous design, so the reply’s argument would reach it, but the reply does not say so. Both tests in the promote_when would need the same expectation control.
- THEORY-458 (Proposed). No bearing.
- QUESTION-048. The exchange is the question in miniature. The reply’s position is that report and implicit measures alike can be produced by demand, and that the only way through is a control that matches expectation; the comment’s is that a perceptual illusion “should be resistant to suggestions”. Both agree the reports occur; the disagreement is over what makes them evidence of an experience. The reply’s own two-routes proposal makes high-suggestibles’ reports genuine experience and low-suggestibles’ bias, which no measure in either paper distinguishes.
- NOTE-372, claim C6. The reply leaves “body ownership can be manipulated by multisensory conflict” undecided: reports of ownership can be manipulated by the rubber-hand procedure, and whether the multisensory conflict is what does it is, on this reply, not yet shown.
- No instruction for machine-learning practice.
Limitations
- A reply, post hoc on both sides by its own account. The expectancy and ownership regressions are exploratory.
- The size of the expectancy difference, which is the whole argument, is not given here; it is referred to the target article.
- The response-bias route for low-suggestibles is asserted. No measure in the exchange separates generated experience from bias, so the two-routes account fits any pattern of reports.
- Small reporting slips: the ownership SE printed as 1.0; the expectancy slope negative with a positive t; Figure 1’s caption does not name the condition.
- The temporal-window argument is not answered directly.
- Seth’s earlier work relies on the design the reply criticises; the reply does not discuss its own authors’ earlier findings, and need not, but a reader of this record should know.
Open questions
- Does the synchrony effect survive when expectations are equalised, for example by telling participants, convincingly, that the asynchronous condition is the illusion condition? The target study tried and its instructions did not move expectancy; a stronger manipulation would settle the main point.
- Do participants expect the illusion to fade at 200–300 ms of asynchrony and not before? If not, the temporal window is evidence the demand account must explain.
- Is the ownership item’s dependence on hypnotisability also present in the asynchronous condition, as the comment reports for the mean of the three items? The reply’s Figure 1 does not say.